My MSI RTX 5070 VENTUS 3X OC intermittently turns both displays black during relatively light or variable workloads such as Netflix, CS2, and Discord. The computer itself appears to remain active: audio continues, Discord users can still hear me, and keyboard sounds still play. The GPU reset shortcut does nothing, so I have to hold the power button to recover.
The system is stable during 15–20 minutes of combined GPU, CPU, and RAM stress testing, even with CS2 running. Temperatures and voltage readings also look normal. During one Netflix-related display blink, the GPU was using about 30 W while its clock moved between roughly 585 and 1807 MHz. The 12VHPWR reading was 12.05 V and the PSU 12V rail was 12.10 V.
Event Viewer records Kernel-Power 41 with BugcheckCode 0, with occasional nvlddmkm Event 153 entries and some earlier LiveKernelEvent 141/117 reports. There are no WHEA errors. I have updated the motherboard BIOS and NVIDIA drivers, disabled overlays and ShadowPlay, reseated the power connector, disabled EXPO, and tested with RAM at DDR5-4800. The system uses a Ryzen 5 7600X, MSI B650 Gaming Plus WiFi, 32 GB DDR5, an MSI MPG A850GF 850 W PSU, Windows 11 24H2, and a 3440x1440 164 Hz DisplayPort monitor alongside a 1080p HDMI display.
Has anyone with a similar RTX 5070 experienced this, and found a stable driver or reliable workaround? I am also wondering whether forcing maximum performance, reducing the core clock, or locking a voltage/frequency curve can prevent failures caused by aggressive low-power state changes.
3 Answers
Do not completely rule out the PSU just because the displayed 12V readings are stable. A failing supply can produce noise or behave badly during rapid load and power-state changes, while still looking fine in software and passing a heavy stress test. A replacement PSU or a temporary test with another known-good unit would be a worthwhile isolation step, especially if the black screens disappear immediately afterward.
The 141/117 and nvlddmkm entries point toward a display-driver timeout or GPU/display-path failure, even though the machine does not fully crash. Try a clean driver installation with a known-good rollback version, then test with one monitor at a time and temporarily lower the primary display refresh rate. If the issue continues across several driver versions, different display configurations, and a separate PSU, the graphics card itself becomes the strongest RMA candidate.
This pattern can be related to display-driver transitions rather than sustained GPU load. Similar cases have been triggered when Netflix advances to the next episode or when video playback changes states. Potential isolation tests include disabling hardware acceleration, trying a different browser, changing the browser's ANGLE graphics backend to DX11 or Vulkan, testing the Windows video app, and disabling Multiplane Overlay. These steps do not guarantee a fix, but they can help identify whether the failure occurs in the browser/video compositor path instead of during 3D rendering.
That is useful because the automatic Netflix episode transition is one of the most repeatable triggers. I have already removed the NVIDIA overlay, but I will test the browser rendering and MPO settings separately.
There are no dwm.exe errors, and the PSU is less than a year old, so I initially considered it unlikely. Still, the light-load behavior could make a PSU swap useful as a final hardware test.